[8.x] [Security GenAI][BUG] KB index entry created via pdf upload does not give the right response (#198020) #198076
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…give the right response (elastic#198020) ## Summary These changes fix the issue with the wrong response of the AI Assistant using knowledge base tool and index entry generated from a PDF file. The issue happens because we are using the first chunk of uploaded PDF document as a context that we pass to LLM instead of using inner hits chunks which are actual parts of the document relevant to the questions. Here is [the blog post](https://www.elastic.co/search-labs/blog/semantic-text-with-amazon-bedrock) that talks about the strategy of using inner hits to get the most relevant documents. (see `Strategy 1: API Calls` section) ### Upload + index PDF 1. Navigate to Integrations page 2. Select "Upload a file" 3. Select and upload a PDF file 4. Press Import button 5. Switch to Advanced tab 6. Fill in "Index name" 7. Add additional field > Add semantic text field > Fill in form * Field: `attachment.content` * Copy to field: `content` * Inference endpoint: `elser_model_2` 8. Press Add button 9. Press Import button ### Add KB index entry (with uploaded PDF data) 1. Navigate to AI Assistant's Knowledge Base page 2. New > Index 3. Fill in "New index entry" form (below are main fields) * Name: `[add entry name]` * Index: `[select index name created during uploading a PDF file]` * Field: `content` 4. Press Save button ### Testing notes Enable knowledge base feature via ``` xpack.securitySolution.enableExperimental: - 'assistantKnowledgeBaseByDefault' ``` ### Example PDF for testing **PDF document**: [Elastic Global Threat Report 2024](https://github.com/user-attachments/files/17544720/elastic-global-threat-report-2024.pdf) **KB Index entry**: Data Description: "Use this tool to answer questions about the Elastic Global Threat Report (GTR) 2024" Query Instruction: "Key terms to return data relevant to the Elastic Global Threat Report (GTR) 2024" **Questions**: 1. Who are the authors of the GTR 2024? 2. What is the forecast for the coming year in GTR 2024? 3. What are top 10 Process Injection by rules in Windows endpoints in GTR 2024? 4. What is the most widely adopted cloud service provider this year according to GTR 2024? 6. Give a brief conclusion of the GTR 2024 **Current behaviour**: <img width="656" alt="Screenshot 2024-10-28 at 16 43 48" src="https://github.com/user-attachments/assets/90615356-8807-4786-b58d-ca28c83aaec9"> **Fixed behaviour**: <img width="655" alt="Screenshot 2024-10-28 at 16 44 47" src="https://github.com/user-attachments/assets/9ebefbcc-20c2-4c79-98f3-11fa6acf3da6"> (cherry picked from commit af2bff4)
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
💚 Build Succeeded
Metrics [docs]
cc @e40pud |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Backport
This will backport the following commits from
mainto8.x:Questions ?
Please refer to the Backport tool documentation